Archive - Joe Dong | Physical AI Deep Dives

System Thinking and Model Thinking in Robot Learning

Both schools are betting on the same variable -- DATA.

May 11 • Joe Dong

April 2026

Robot Learning Should Be Goal-Driven and Data-First

A goal-driven view on robot learning, industrial deployment, and the data engine needed to scale dexterous robots.

Apr 28 • Joe Dong

π0.7: Prompt Engineering Comes to Robot Foundation Models

π0.7’s key advance is full action context, not just a stronger model—unlocking compositional generalization.

Apr 23 • Joe Dong

NVIDIA EgoScale: Pretraining Dexterous Manipulation with 20,000 Hours of Egocentric Human Data

A simple but important idea: pretrain on large-scale human egocentric data first, then use a small amount of aligned human–robot data and robot…

Apr 14 • Joe Dong

Nvidia DreamZero0: Why World Models May Become Policies

DreamZero0 is not just a stronger robotics benchmark.

Apr 8 • Joe Dong

#nojs-banner { position: fixed; bottom: 0; left: 0; padding: 16px 16px 16px 32px; width: 100%; box-sizing: border-box; background: red; color: white; font-family: -apple-system, "Segoe UI", Roboto, Helvetica, Arial, sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol"; font-size: 13px; line-height: 13px; } #nojs-banner a { color: inherit; text-decoration: underline; } This site requires JavaScript to run correctly. Please turn on JavaScript or unblock scripts